The Transformation of Knowledge Work: How AI Agents Like Codex and GPT-5.5 Are Revolutionizing Developer Workflows

AI agents have drastically transformed the way developers work, and their influence is beginning to extend into the realm of knowledge work. This includes processing information, solving intricate problems, generating new ideas, and driving innovation. One standout player in this emerging field is Codex, OpenAI’s cutting-edge coding application, now powered by the advanced GPT-5.5 model.

The Power of GPT-5.5

OpenAI’s latest model, GPT-5.5, integrates seamlessly with Codex and operates on NVIDIA’s GB200 NVL72 rack-scale systems. These systems offer impressive capabilities, such as delivering 35x lower cost per million tokens and 50x higher token output per second per megawatt compared to previous models. This operational efficiency means that utilizing frontier-model inference on a large scale is now feasible for enterprises.

A Widespread Adoption Among NVIDIANs

As a testament to its efficacy, over 10,000 employees at NVIDIA—from engineering to HR—are already leveraging GPT-5.5-powered Codex in their daily workflows. Users describe the outcomes as “mind-blowing” and “life-changing,” highlighting the profound impact AI is having on productivity across various departments.

Measurable Gains in Development Cycles

NVIDIA engineers have reported significant improvements after gaining access to GPT-5.5 through Codex. Debugging processes that once spanned days are now reduced to mere hours. Complex experimentation that previously took weeks can be accomplished overnight, especially in intricate, multi-file codebases. Moreover, teams are now effectively generating end-to-end features from simple natural-language prompts, with enhanced reliability and efficiency compared to earlier models.

NVIDIA and OpenAI: A Strategic Partnership

This impressive progress underscores the collaborative efforts between NVIDIA and OpenAI. Their partnership has not only accelerated AI agent utilization within NVIDIA but has also paved the way for other companies to build the most efficient AI models available. As NVIDIA founder and CEO Jensen Huang articulated in a company-wide email, “Let’s jump to lightspeed. Welcome to the age of AI.”

Enterprise Security: A Key Consideration

However, with great power comes great responsibility, especially when navigating secure enterprise environments. Each AI agent requires a dedicated computer to operate effectively and securely. Codex supports remote Secure Shell (SSH) connections to authorized cloud virtual machines (VMs), ensuring that sensitive company data remains protected while agents work.

NVIDIA IT has successfully rolled out cloud VMs for all employees, providing a secure sandbox environment for agents to operate at peak performance. This allows users to maintain full oversight with a familiar interface while controlling the Codex agent from their cloud VM. Additionally, a zero-data retention policy governs the system, and agents only access production environments with read-only permissions through command-line interfaces.

A Decade of Cooperation in AI

The launch of GPT-5.5 and the functionality of Codex are the fruits of more than ten years of collaboration between NVIDIA and OpenAI. This partnership began in 2016 when Huang delivered the first NVIDIA DGX-1 AI supercomputer to OpenAI’s headquarters in San Francisco. Since then, the collaboration has deepened, with NVIDIA serving as a key partner for OpenAI’s developments, including the gpt-oss open-weight model launch.

OpenAI has committed to deploying over 10 gigawatts of NVIDIA systems for future AI infrastructure, establishing a foundation that will leverage millions of NVIDIA GPUs for model training and inference for years to come.

Crucially, NVIDIA and OpenAI enjoy a relationship of mutual influence, where feedback from OpenAI helps to shape NVIDIA’s hardware plans, while OpenAI gains early access to upcoming architectures. This collaborative spirit has led to impressive milestones, including the successful integration of the GB200 NVL72 100,000-GPU cluster, a model that has set new benchmarks in system reliability and performance.

Overall, GPT-5.5 embodies the capabilities of this robust infrastructure, showcasing the potential of AI agents to redefine workflows and elevate productivity to unprecedented levels.